Motion analysis via interframe point correspondence establishment

摘要

The matter of feature correspondence is essential for motion analysis. As the most general case of motion, we consider the case of particle tracking in fluid. Particles are treated as single points of their centroids. The problem of the point correspondence is stated as a total path optimization task, solved by using the dynamic programming technique. This new global approach is more promising than the local techniques that have been used so far. The pairing consistency is additionally checked via a statistical test. The method is suitable for rigid motion tracking as well. Preliminary experiments both with fluid and solid body deformable surfaces have proven the method's flexibility and generality. The computational expenses of the algorithm are estimated, and the performance results for 300–400 particle tracking tasks are clearly promising.